​
Installation

The package is available from the public PyPi. Install it using the following command:

pip install --upgrade cerebrium

If you have trouble installing the package, please contact the team immediately!

​
Set up your Cerebrium account

  1. If you haven’t already, sign up for Cerebrium using your email address or Google login credentials.
  2. From your dashboard you will be able to view all deployed models, API calls, versions and invite teammates!

The first thing you will need is your private API key for the project that you intend to deploy your model into. You can find the API key by navigating to the API keys section in your dashboard on Cerebrium which should look as follows:

private_api_key Note that each project has a unique API key so double check you are using the key that corresponds to the right project.

Secondly, you will need to make sure that you have the latest version of the CLI installed and on your path. The CLI is provided by the Cerebrium python package, so if you have not installed that yet, go ahead and do so by running pip install cerebrium.
If you have Cerebrium installed already, we recommend you upgrade your package with pip install --upgrade cerebrium as we are frequently releasing new features and improvements!

You can verify your installation by running cerebrium --help in your terminal.
If you do not get any output but you can run import cerebrium` in Python, ensure pip is on your path.

​
Login Usage

To use the login feature, navigate to your terminal window which you would like to use to deploy your models and run the following command:

cerebrium login <YOUR_PRIVATE_API_KEY>

This will store your API key locally, ensuring that you are logged in whenever you open a terminal and use cerebrium deploy without you having to enter your API key again.

Some important details to note:

  • You will remain logged in between different terminals provided you are the same user.
  • You can still parse in an API key when deploying your models without being logged out. The parsed API key takes preference over the local one. This is particularly useful if you are deploying to a different project.
  • If you would like to remove your login details, remove the entry in ~/.cerebrium/config.yaml

Awesome! Let’s start creating ML-based applications

​
How to navigate our documentation

Cerebrium caters to many different types of businesses and users so we have created abstractions that suit different users in different situations. Based on your use-case we recommend you look at the starting points below and examples in each sub-directory.

​
When would you use Conduit over Cortex

  • The conduit is a very high level abstraction for deploying ML models and gives the end user a very streamlined and easy way to deploy their ML models. Its best suited if you would like to deploy a model file or a HuggingFace model ID to production quickly with minimal setup. We recommend this for simple use cases that have simple pre/post processing or logging logic.But don’t let the simplicity fool you, it can still scale to 10k requests per second.
  • We recommend using Cortex if you would like fine-grain control over your environment, package versions and implementation. Typically our clients use this when they have more complex logic, stricter latency requirements using packages such as Deepspeed, TensorRT etc and are communicating with many external tools.

Conduit

You have a model file or Hugging Face model ID and just want to turn it into an API endpoint.

Cortex

Deploy any custom Python code. Use this if you want very fine-grain control.

Prebuilt models

You would like to deploy popular existing models with one click.

IntroductionSecurity